1 downloads
0 Views
613KB Size

Letter pubs.acs.org/JPCL

Practical Density Functionals beyond the Overdelocalization− Underbinding Zero-Sum Game Benjamin G. Janesko,*,† Emil Proynov,† Jing Kong,‡ Giovanni Scalmani,¶ and Michael J. Frisch¶ †

Department of Chemistry, Texas Christian University, Fort Worth, Texas 76129, United States Department of Chemistry, Middle Tennessee State University, Murfreesboro, Tennessee 37132, United States ¶ Gaussian, Inc., Wallingford, Connecticut 06492, United States ‡

S Supporting Information *

ABSTRACT: Density functional theory (DFT) uses a density functional approximation (DFA) to add electron correlation to mean-ﬁeld electronic structure calculations. Standard strategies (generalized gradient approximations GGAs, meta-GGAs, hybrids, etc.) for building DFAs, no matter whether based on exact constraints or empirical parametrization, all face a zero-sum game between overdelocalization (fractional charge error, FC) and underestimation of covalent bonding (fractional spin error, FS). This work presents an alternative strategy. Practical “Rung 3.5” ingredients are used to implement insights from hyper-GGA DFAs that reduce both FS and FC errors. Prototypes of this strategy qualitatively improve FS and FC error over 40 years of standard DFAs while maintaining low cost and practical evaluation of properties. Numerical results ranging from transition metal thermochemistry to absorbance peaks and excited-state geometry optimizations highlight this strategy’s promise and indicate areas requiring further development.

D

Unfortunately, because exact exchange exactly cancels the Hartree self-interaction, approximating exchange makes each electron interact with itself. This causes a litany of overdelocalization errors in atomic cores,18 Rydberg states,19 paramagnetic defects,20,21 insulators,22 charge-transfer excitations,23 conical intersections,24,25 many transition states,26 and FC atoms as in dissociated H+2 . Figure 1 shows how 40 years of standard DFAs all suﬀer the same zero-sum trade-oﬀ between FS and FC error. DFAs giving low FS error have relatively high FC error, and vice versa. Despite substantial recent progress in other areas,27−29 users can be forgiven for regarding most new DFAs as little better than the old ones. Figure 1 also highlights DFT developers’ two-decade project to address this crisis: using a new ingredient to do by design what approximate exchange does accidentally.30−40 The hypergeneralized gradient approximations (hyper-GGAs) in Figure 1 use the exact exchange energy density30 to build better approximations for correlation.

ensity functional theory (DFT) is arguably experiencing a crisis of conﬁdence.1 As developers devise new density functional approximations (DFAs) based on old strategies (meta-GGAs, global and range-separated hybrids; Figure 1), users (including the authors2) often stick to3 DFAs that are over 20 years old.4−6 Vigorous debates over developers “straying from the path”7−9 and spiraling eﬀorts to add expensive ab initio corrections1 highlight this turmoil. Figure 1 helps explain why users often use older DFAs. Kohn−Sham DFT10 corrects mean ﬁeld (molecular orbital, Hartree−Fock-like) calculations with a DFA for many-body correlation. These corrections should ensure, for example, that the energy of dissociated ground-state singlet H2 does not include Hartree−Fock theory’s contributions from high-energy ionic conﬁgurations H+···H− and H−···H+ placing both electrons on the same atom (Figure SI-1). Standard strategies for building DFAs, whether designed solely around exact constraints11 or incorporating empirical ﬁts,12 also replace some or all of the exact exchange with a semilocal approximation. This surprising strategy to “contaminate” an exact quantity with an approximate one fortuitously approximates the correlations that, for example, localize dissociated H2’s electrons to diﬀerent atoms.13 Approximating correlation with exchange reduces this fractional spin (FS) error and is at the heart of DFT’s useful accuracy for covalent bonds13 and metals.14 (Each H atom in dissociated H2 has, on average, 1/2 ↑-spin and 1/2 ↓-spin electrons; hence “fractional spin”; refs 15 and 16 provide illuminating perspectives on FS and fractional charge (FC) errors and show how the model systems in Figure 1 isolate these errors’ eﬀects. Symmetry breaking also reduces FS error but is not a panacea.17) © 2017 American Chemical Society

eXσ ( r ⃗) = −

1 2

∫ d3r ′⃗

|γσ( r ⃗ , r ′⃗ )|2 | r ⃗ − r ′⃗ |

(1)

(Here γσ(r,⃗ r′⃗ ) = ∑iσ ψiσ(r)⃗ ψ*iσ(r′⃗ ) is the σ = ↑,↓-spin oneparticle density matrix of the noninteracting Kohn−Sham reference system, made from occupied MOs {ψiσ(r)⃗ }. We use hyper-GGA to denote a position-dependent admixture of the exact-exchange energy density,31 not the standard hybrid Received: August 2, 2017 Accepted: August 24, 2017 Published: August 24, 2017 4314

DOI: 10.1021/acs.jpclett.7b02023 J. Phys. Chem. Lett. 2017, 8, 4314−4318

Letter

The Journal of Physical Chemistry Letters eX̃ σ ( r ⃗) = (aNdσ ( r ⃗))2/3 eXPBE σ ( r ⃗)

Ndσ ( r ⃗) = ρσ−1( r ⃗)|

(3)

∫ d3r ′⃗ γσ( r ⃗ , r ′⃗ )ϕr1s⃗;d( r ′⃗ )|2

f ( r ⃗) = (1 + b(s 2 + (d /rs)4 + (rs/d)4 ))−1

(4) (5)

We are inspired by Becke’s use of eq 1 to estimate the fraction of the exact exchange hole hXσ(r,⃗ r′⃗ ) = −ρσ−1(r)⃗ |γσ(r,⃗ r′⃗ )|2 localized around point r.⃗ Delocalization of both hX↑ and hX↓ indicates potential FS error.32,33 The key Rung 3.5 quantity Ndσ(r)⃗ (eq 4) projects γσ onto atom-like test functions centered at each point r,⃗ ϕ1s r⃗;d(r′⃗ ) =

admixture of integrated exact exchange.) Becke’s seminal realspace nondynamical correlation functional B05,32,33 the KP16/ B13 reﬁnement,34,35 and PSTS31 all qualitatively improve over standard DFAs, within a practical occupied-orbital-only framework.1 Several recent Letters to this journal explicitly address aspects of this important problem.24,25,41,42 Unfortunately, hyper-GGAs are today almost exclusively used by hyper-GGA developers. Evaluating eq 1 is more computationally expensive than evaluating integrated exact exchange. Analytically diﬀerentiating eq 1 is even more expensive,43 and such derivatives (with respect to nuclear position, nuclear magnetic moment, applied electromagnetic ﬁelds, etc.) enable practical predictions of bond lengths, IR/ Raman and NMR spectra, excited-state geometries and ﬂuorescence, and other properties that users care about.44 The eﬃciency of evaluating eq 1 and its derivatives has improved recently, starting from approximate schemes based on the resolution of identity36,37 to more recent exact seminumerical schemes.38−40 The latter oﬀer eﬃciency and precision advantageous for large systems39,40 but require additional coding and implementation eﬀort. Substantial development eﬀorts are needed before hyper-GGAs are available for routine application to chemistry. We propose an alternative strategy: devise other new ingredients that capture the physics of eq 1 at reduced cost and use these ingredients to implement the insights of hyperGGAs.30−40 We present the ﬁrst major steps down this road, two prototype DFAs HF-FS and PBEHH-FS. These combine exact exchange (HF) or the half-and-half global hybrid47 of exact and PBE exchange (PBEHH) with a size-consistent, analytically diﬀerentiable “Rung 3.5” (ref 48) correlation functional correcting FS error.

∑ ∫ d3r ⃗ f ( r ⃗)(1 − F[ζ ])(eXPBE σ ( r ⃗) − eX̃ σ ( r ⃗)) σ

3/4

exp( −| r ⃗ − r ′⃗ |2 /d 2), to

determine the extent to which the density matrix itself is localized within distance d about point r.⃗ (DFT+U uses analogous projections onto atomic states centered at nuclei.49) Figure 2 shows that for d on the order of 1−2 bohr, Ndσ(r)⃗ is

Figure 1. FC vs FS errors. Open diamonds are 40 years of standard DFAs: APFD, B3LYP, B971, B972, B97D, BHHLYP, BLYP, BMK, BRxBRc, CAMB3LYP, HCTH147, HCTH407, HCTH93, Slater exchange, HF, HF plus LDA/PBE/TPSS correlation, HISS, HSE, LCwPBE, LDA, M06, M062X, M06HF, M06L, M08HX, M11, MN12SX, MN12L, MN15, MN15L, N12-SX, O3LYP, PBE, PW6B95, PW91, SOGGA11X, TPSS, VSXC, wB97XD, X3LYP, and Xα. Filled diamonds are hyper-GGAs. Filled circles are simpliﬁed hyper-GGAs with the single empirical parameter b = 0.05 (FS05) or thermochemically optimal b (FSopt).

ECFS =

2 πd 2

( )

Figure 2. Nd↑(r⃗) in the H atom (red solid) and a H in stretched symmetric singlet H2 (blue dashed); d = 2.21 bohr. (Inset) Similar results for the N atom and N2; d = 1 bohr.

small in density tails and atomic cores, relatively large in the integer-occupied valence regions of isolated H and N atoms, and smaller in the fractionally occupied valence regions of stretched symmetric singlet H2 and N2. Reference 50 presents a more extensive study of the electron delocalization range function, whose square equals N dσ (r )⃗ . ẽX σ (r )⃗ (eq 3) approximates eq 1 by scaling down PBE exchange in fractionally occupied valence regions where Ndσ(r)⃗ is relatively PBE small. EFS C (eq 2) uses eXσ − ẽXσ, approximating the diﬀerence between exact and PBE exchange, as an eﬀective opposite-spin correlation energy density. This is added to spin-unpolarized regions where F[ζ] = [(1 + ζ)4/3 + (1 − ζ)4/3 − 2]/(24/3 − 2),51 ζ = (ρ↑(r)⃗ − ρ↓(r)⃗ )/ρ(r)⃗ , is near zero. Because Ndσ(r)⃗ only provides useful information where rs = (4πρ(r)⃗ /3)−1/3 is close to distance d and reduced gradient s = |∇ρ(r)⃗ |/(2kFρ), kF = (6π2ρ)1/3, is not too large (Figure 2), the “mixing function” (eq 5) goes to zero outside of such regions. This ensures EFS C → 0 in low- and high-density regions rs ≫ d and rs ≪ d, respectively an exact constraint and a reasonable choice.31 EFS C contains three parameters d,a,b. d = 2.21 bohr is ﬁt to “appropriate norms”,27 averaging the overlap distances52 (d maximizing system-averaged Nd(r)⃗ ) of the H atom and He atom. a = 1.57 ensures ∫ d3r ⃗ ẽX↑(r)⃗ equals the exact exchange energy of the appropriate norm H atom. b is an empirical parameter. The HF-FS exchange−correlation (XC) energy is PBE EX + EFS C + EC , the latter being the PBE dynamical correlation

(2) 4315

DOI: 10.1021/acs.jpclett.7b02023 J. Phys. Chem. Lett. 2017, 8, 4314−4318

Letter

The Journal of Physical Chemistry Letters Table 1. Errors in Ground-State Properties MGAEc

ccCA-TM/11c

HTd

NHTd

method

Etota

b

BL

RMS

max

RMS

max

ΔE

Barr

ΔE

Barr

Ze

Rotf

HF-PBE HF-FSopt PBEHH PBEHH-FSopt PBE0 PBE

−0.9 14.5 −4.6 1.6 −6.3 −7.9

4.20 5.84 2.13 2.56 0.99 1.09

38.3 24.2 14.0 13.0 6.2 18.0

89g 59h 35i 34h 22i 47j

139.5 132.2 68.5 66.3 23.0 39.2

749k 828l 433l 373k 34k 179l

4.1 3.8 1.7 2.2 1.1 1.8

7.8 11.0 1.8 2.8 4.8 10.4

8.4 10.5 4.6 5.3 4.5 6.4

10.2 16.9 4.2 5.5 6.8 13.1

−6.7 −1.1 −2.8 0.2 −0.6 1.6

−0.41 −0.26 −0.11 −0.04 0.04 0.18

a

Mean error (ME) per electron (mH, accurate-predicted) in atom H−Ar total energies. bRoot-mean-square (RMS) deviation (pm) in MGBL20 bond lengths. cRMS and max errors (kcal/mol) in MGAE109 atomization energies and ccCA-TM/11 3d transition metal complex ΔHf°. dRMS (kcal/mol) in (N)HTBH38/08 reaction energies ΔE and barriers Barr. eError (10−2 e) in H2 + H transition state transferred atom spin polarization Z, CCSD reference value −9.3 × 10−2 e (ref 45); negative error denotes overpolarization. fError in ethylene rotational barrier Rot (eV), experimental value 2.59 eV (ref 46). gCF3CN. hSpiro-C5H8. iSiF4. jC5H5N. kCr3O9. lV4O10.

functional.4 The PBEHH-FS XC energy is 1 1 E + 2 E XPBE + ECFS + ECPBE . B05, PSTS, and KP16/B13 are 2 X calculated self-consistently as dissociated atoms using an inhouse program in Kong’s group. All other generalized Kohn− Sham calculations use the development version of the Gaussian suite of programs.53 Other computational details are in ref 54 and the Supporting Information. Figure 1 shows that both DFAs qualitatively improve over 40 years of standard approximations. Choosing the single empirical parameter (compare to ﬁve in PSTS31) b = 0.05 highlights the tremendous advances possible, while thermochemically optimal b = 0.15 (HF-FSopt) and b = 0.5 (PBEHH-FSopt) provide more general improvements (Table 1). Reducing b increases the magnitude of FS correction. (Figure SI-1 conﬁrms that HFFSopt and PBEHH-FSopt are identical to the underlying HFPBE and PBEHH DFAs for H+2 dissocation and improve on the underlying DFAs for the entire H2 dissociation potential energy surface.) FS correction improves the underlying DFAs’ maingroup and 3d transition metal thermochemistry, shortens covalent bonds (Table SI-3), maintains reasonable treatments of bond lengths and radical reaction barriers, and reduces the density-driven error55 in H2 + H transition state polarization and the error in symmetry-broken singlet calculations on the ethylene rotational barrier. Hyper-GGAs are rarely used to simulate spectra due to the diﬃculty of diﬀerentiating eq 1.43 Table 2 reports adiabatic linear response time-dependent (TD-)DFT calculations56 on the absorption energy of the He atom and a lithium ﬂuoride Fcenter (color center)20 and the geometry and adiabatic transition energy Te of the formaldehyde 1A″ excited

state.44,57 FS corrections improve the overestimated HF-PBE helium and F-center excitations and the overestimated HF-PBE and PBEHH formaldehyde Te while giving only modest changes in the formaldehyde excited-state geometry. The computational cost of standard hyper-GGAs is at least that of global hybrids due to the extra cost of evaluating eq 1. Table 3 shows that FS correlation can have a cost below global Table 3. Computational Timings in Polyalaninea PBE HF exchangeb FS correlationc FS,c d = 1 bohr FS,c d = 0.5 bohr a

Total wall times (minutes) for a single SCF cycle on extended alanine25 peptide, single-processor calculation, 6-311+G(2d,p) basis set. bHFPBE time minus PBE time. cHF-FS time minus HFPBE time.

hybrid exchange in a relatively large peptide. Decreasing d accelerates evaluation by reducing the range of the integrand in eq 4. These results are a major step toward practical DFAs beyond the overdelocalization−underbinding zero-sum game. However, the journey is not complete. Equation 2 does not outperform standard DFAs for thermochemistry or radical reaction barriers, even when “contaminated” with 50% approximate exchange. (For comparison, the KP16/B13 hyper-GGA does not outperform B3LYP for atomization energies.35) The remaining task is to incorporate eq 4 and related ingredients into more sophisticated replacements for eq 2, designed around exact constraints11 and potentially including empirical ﬁts.12 We believe that further development of this practical, physically motivated, analytically diﬀerentiable, occupied-orbital-only1 strategy has the promise to improve over DFAs (Figure 1) constructed using standard strategies.

Table 2. Predictions of Excited-State Properties method

Hea

F-cenb

R(CO)c

ϕc

Tce

Ref HF-PBE HF-FSopt PBEHH PBEHH-FSopt PBE0 PBE

21.2 25.19 23.74 23.65 22.92 22.89 22.13

5.08 5.43 5.37 5.05 5.00 4.79 4.46

1.32 1.233 1.222 1.266 1.264 1.285 1.308

16.7 17.4 18.9 19.2 20.2 21.7

3.49 4.33 4.10 3.85 3.71 3.62 3.39

47 136 69 41 24

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jpclett.7b02023. Details of calculations and test sets, numerical evaluation of Ndσ(r)⃗ , total energies from Figure 1, citations to DFAs in Figure 1 and test sets in Table 1, excited-state geometries from Table 2, total energies and wall times from Table 3, H2 and H+2 potential energy surfaces, and plots of Ndσ(r)⃗ in representative systems(PDF)

a First bright state excitation energy of the He atom (eV). bFirst bright state excitation energy of the F-center defect in crystalline lithium ﬂuoride (eV). cOptimized CO bond length (Angstrom), out-ofplane angle ϕ (degrees), and adiabatic transition energy Te (eV) of the formaldehyde 1 1A″ excited state.

4316

DOI: 10.1021/acs.jpclett.7b02023 J. Phys. Chem. Lett. 2017, 8, 4314−4318

Letter

The Journal of Physical Chemistry Letters

■

(19) Caricato, M.; Trucks, G. W.; Frisch, M. J.; Wiberg, K. B. Electronic Transition Energies: Study of the Performance of Large Range of Single Reference Density Functional and Wave Function Methods on Valence and Rydberg States Compared to Experiment. J. Chem. Theory Comput. 2010, 6, 370. (20) Karsai, F.; Tiwald, P.; Laskowski, R.; Tran, F.; Koller, D.; Gräfe, S.; Burgdörfer, J.; Wirtz, L.; Blaha, P. F center in lithium fluoride revisited: Comparison of solid-state physics and quantum-chemistry approaches. Phys. Rev. B: Condens. Matter Mater. Phys. 2014, 89, 125429. (21) Laegsgaard, J.; Stokbro, K. Hole Trapping at Al impurities in Silica: A Challenge for Density Functional Theories. Phys. Rev. Lett. 2001, 86, 2834−2837. (22) Mori-Sánchez, P.; Cohen, A. J.; Yang, W. Localization and Delocalization Errors in Density Functional Theory and Implications for Band-Gap Prediction. Phys. Rev. Lett. 2008, 100, 146401. (23) Dreuw, A.; Weisman, J. L.; Head-Gordon, M. Long-range charge-transfer excited states in time-dependent density functional theory require non-local exchange. J. Chem. Phys. 2003, 119, 2943. (24) Gao, J.; Grofe, A.; Ren, H.; Bao, P. Beyond Kohn-Sham Approximation: Hybrid Multistate Wave Function and Density Functional Theory. J. Phys. Chem. Lett. 2016, 7, 5143−5149. (25) Ren, H.; Provorse, M. R.; Bao, P.; Qu, Z.; Gao, J. Multistate Density Functional Theory for Effective Diabatic Electronic Coupling. J. Phys. Chem. Lett. 2016, 7, 2286−2293. (26) Mahler, A.; Janesko, B. G.; Moncho, S.; Brothers, E. N. Why are GGAs so accurate for reaction kinetics on surfaces? Systematic comparison of hybrid vs. nonhybrid DFT for representative reactions. J. Chem. Phys. 2017, 146, 234103. (27) Sun, J.; Ruzsinszky, A.; Perdew, J. P. Strongly Constrained and Appropriately Normed Semilocal Density Functional. Phys. Rev. Lett. 2015, 115, 036402. (28) Grimme, S. Density functional theory with London dispersion corrections. WIREs Computational Molecular Science 2011, 1, 211−228. (29) Stein, T.; Eisenberg, H.; Kronik, L.; Baer, R. Fundamental gaps in finite systems from eigenvalues of generalized Kohn-Sham method. Phys. Rev. Lett. 2010, 105, 266802. (30) Tao, J.; Staroverov, V. N.; Scuseria, G. E.; Perdew, J. P. Exactexchange energy density in the gauge of semilocal functional. Phys. Rev. A: At., Mol., Opt. Phys. 2008, 77, 012509. (31) Perdew, J. P.; Staroverov, V. N.; Tao, J.; Scuseria, G. E. Density functional with full exact exchange, balanced nonlocality of correlation, and constraint satisfaction. Phys. Rev. A: At., Mol., Opt. Phys. 2008, 78, 052513. (32) Becke, A. D. A real-space model of nondynamical correlation. J. Chem. Phys. 2003, 119, 2972−2977. (33) Becke, A. D. Real-space post-Hartree-Fock correlation models. J. Chem. Phys. 2005, 122, 064101. (34) Becke, A. D. Density functionals for static, dynamical, and strong correlation. J. Chem. Phys. 2013, 138, 074109. (35) Kong, J.; Proynov, E. Density Functional Model for Nondynamic and Strong Correlation. J. Chem. Theory Comput. 2016, 12, 133−143. (36) della Sala, F.; Görling, A. Efficient localized Hartree-Fock methods as effective exact-exchange Kohn-Sham methods for molecules. J. Chem. Phys. 2001, 115, 5718−5732. (37) Proynov, E.; Liu, F.; Shao, Y.; Kong, J. J. Chem. Phys. 2012, 136, 034102. (38) Neese, F.; Wennmohs, F.; Hansen, A.; Becker, U. Efficient, approximate and parallel Hartree-Fock and hybrid DFT calculations. ’chain-of-spheres’ algorithm for the Hartree-Fock exchange. Chem. Phys. 2009, 356, 98. (39) Bahmann, H.; Kaupp, M. Efficient Self-Consistent implementation of local hybrid Functionals. J. Chem. Theory Comput. 2015, 11, 1540. (40) Liu, F.; Kong, J. Efficient Computation of Exchange Energy Density with Gaussian Basis Functions. J. Chem. Theory Comput. 2017, 13, 2571.

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected] ORCID

Benjamin G. Janesko: 0000-0002-2572-5273 Emil Proynov: 0000-0002-1141-241X Jing Kong: 0000-0002-0649-2232 Notes

The authors declare no competing ﬁnancial interest.

■

ACKNOWLEDGMENTS This work was supported by the National Science Foundation DMR-1505343.

■

REFERENCES

(1) Becke, A. D. Perspective: Fifty years of density-functional theory in chemical physics. J. Chem. Phys. 2014, 140, 18A301. (2) Li, L.; Janesko, B. G. 3-Methyleneisoindolin-1-one Assembly via Base- and CuI/l-Proline-Catalyzed Domino Reaction: Mechanism of Regioselective Anionic Cyclization. J. Org. Chem. 2016, 81, 10802− 10808. (3) Burke, K. Perspective on density functional theory. J. Chem. Phys. 2012, 136, 150901. (4) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient Approximation Made Simple. Phys. Rev. Lett. 1996, 77, 3865−3868. (5) Becke, A. D. Density-Functional Thermochemistry. III. The Role of Exact Exchange. J. Chem. Phys. 1993, 98, 5648−5652. (6) Stephens, P. J.; Devlin, F. J.; Chabalowski, C. F.; Frisch, M. J. Ab Initio Calculation of Vibrational Absorption and Circular Dichroism Spectra Using Density Functional Force Fields. J. Phys. Chem. 1994, 98, 11623−11627. (7) Medvedev, M. G.; Bushmarinov, I. S.; Sun, J.; Perdew, J. P.; Lyssenko, K. A. Density functional theory is straying from the path toward the exact functional. Science 2017, 355, 49. (8) Kepp, K. P. Comment on “Density functional theory is straying from the path toward the exact functional”. Science 2017, 356, 496.2. (9) Medvedev, M. G.; Bushmarinov, I. S.; Sun, J.; Perdew, J. P.; Lyssenko, K. A. Response to Comment on “Density functional theory is straying from the path toward the exact functional”. Science 2017, 356, 496.3. (10) Kohn, W.; Sham, L. Self-consistent equations including exchange and correlation effects. Phys. Rev. 1965, 140, A1133−A1138. (11) Tao, J.; Perdew, J. P.; Staroverov, V. N.; Scuseria, G. E. Climbing the Density Functional Ladder: Nonempirical Meta-Generalized Gradient Approximation Designed for Molecules and Solids. Phys. Rev. Lett. 2003, 91, 146401. (12) Peverati, R.; Truhlar, D. G. Quest for a universal density functional: the accuracy of density functionals across a broad spectrum of databases in chemistry and physics. Philos. Trans. R. Soc., A 2014, 372, 20120476. (13) Cremer, D. Density functional theory: coverage of dynamic and non-dynamic electron correlation effects. Mol. Phys. 2001, 99, 1899− 1940. (14) Paier, J.; Marsman, M.; Kresse, G. Why does the B3LYP hybrid functional fail for metals? J. Chem. Phys. 2007, 127, 024103. (15) Cohen, A. J.; Mori-Sánchez, P.; Yang, W. Insights into current limitations of density functional theory. Science 2008, 321, 792−794. (16) Cohen, A. J.; Mori-Sánchez, P.; Yang, W. Challenges for density functional theory. Chem. Rev. 2012, 112, 289−320. (17) Sherrill, C. D.; Lee, M. S.; Head-Gordon, M. On the performance of density functional theory for symmetry-breaking problems. Chem. Phys. Lett. 1999, 302, 425−430. (18) Arbuznikov, A. V.; Kaupp, M. Nuclear shielding constants from localized local hybrid exchange-correlation potentials. Chem. Phys. Lett. 2007, 442, 496. 4317

DOI: 10.1021/acs.jpclett.7b02023 J. Phys. Chem. Lett. 2017, 8, 4314−4318

Letter

The Journal of Physical Chemistry Letters (41) Vuckovic, S.; Gori-Giorgi, P. Simple Fully Nonlocal Density Functionals for Electronic Repulsion Energy. J. Phys. Chem. Lett. 2017, 8, 2799−2805. (42) Kong, J.; Proynov, E.; Yu, J.; Pachter, R. Describing a Strongly Correlated Model System with Density Functional Theory. J. Phys. Chem. Lett. 2017, 8, 3142−3146. (43) Maier, T. M.; Bahmann, H.; Kaupp, M. Efficient Semi-numerical Implementation of Global and Local Hybrid Functionals for TimeDependent Density Functional Theory. J. Chem. Theory Comput. 2015, 11, 4226−4237. (44) Jacquemin, D.; Perpéte, E. A.; Scalmani, G.; Frisch, M. J.; Ciofini, I.; Adamo, C. Absorption and emission spectra in gas-phase and solution using TD-DFT: Formaldehyde and benzene as case studies. Chem. Phys. Lett. 2006, 421, 272−276. (45) Janesko, B. G. Reducing density-driven error without exact exchange. Phys. Chem. Chem. Phys. 2017, 19, 4793−4801. (46) Krylov, A. I.; Sherrill, C. D. Perturbative corrections to the equation-of-motion spin-flip self-consistent field model: Application to bond-breaking and equilibrium properties of diradicals. J. Chem. Phys. 2002, 116, 3194−3203. (47) Becke, A. D. A New Mixing of Hartee-Fock and Local DensityFunctional Theories. J. Chem. Phys. 1993, 98, 1372−1377. (48) Janesko, B. G. Rung 3.5 density functionals. J. Chem. Phys. 2010, 133, 104103. (49) Gani, T. Z. H.; Kulik, H. J. Where Does the Density Localize? Convergent Behavior for Global Hybrids, Range Separation, and DFT +U. J. Chem. Theory Comput. 2016, 12, 5931−5945. (50) Janesko, B. G.; Scalmani, G.; Frisch, M. J. How Far Do Electrons Delocalize? J. Chem. Phys. 2014, 141, 144104. (51) Perdew, J. P. In Electronic Structure of Solids ’91; Ziesche, P., Eschrig, H., Eds.; Akademie Verlag: Berlin, Germany, 1991; pp 11−20. (52) Mehmood, A.; Janesko, B. G. The electron delocalization range in stretched bonds. Int. J. Quantum Chem. 2016, 116, 1783−1795. (53) Frisch, M. J., Trucks, G. W., Schlegel, H. B.; et al. Gaussian Development Version, revision H.10; Gaussian, Inc.: Wallingford, CT, 2010. (54) Janesko, B. G.; Scalmani, G.; Frisch, M. J. Practical auxiliary basis implementation of Rung 3.5 functionals. J. Chem. Phys. 2014, 141, 034103/1−034103/14. (55) Kim, M.-C.; Sim, E.; Burke, K. Understanding and Reducing Errors in Density Functional Calculations. Phys. Rev. Lett. 2013, 111, 073003. (56) Maitra, N. T. Perspective: Fundamental aspects of timedependent density functional theory. J. Chem. Phys. 2016, 144, 220901. (57) Furche, F.; Ahlrichs, R. Adiabatic time-dependent density functional methods for excited state properties. J. Chem. Phys. 2002, 117, 7433−7447.

4318

DOI: 10.1021/acs.jpclett.7b02023 J. Phys. Chem. Lett. 2017, 8, 4314−4318